Actual and Perceived Cues to Deception: A Closer Look at ...sarahita/SLP/depaulo.pdfPrevious research has shown both that speech can reliably reveal whether or not deception is occurring

BASIC AND APPLIED SOCIAL PSYCHOLOGY, 1982, 3 (4) 291-312

Actual and Perceived Cues toDeception: A Closer Look at

Speech

Bella M. DePauloUniversity of Virginia

Robert RosenthalHarvard University

Judith RosenkrantzUniversity of California, Davis

Carolyn Rieder GreenHarvard University

Previous research has shown both that speech can reliably reveal whether ornot deception is occurring and that perceivers are often strongly influenced byspeech in their judgments about deceit. Nonetheless, there are relatively fewstudies of verbal cues to deceit. In the present study, we examined specific ver-bal and paralinguistic cues that might reveal when deception is occurring orthat might be used by perceivers in their attempts to detect deception; also, weexamined quantitatively the correspondence between actuEil cues to deceptionand perceived cues to deception. For the cues that we studied, the degree towhich the cues actually were associated with deception correspondedsignificantly to the degree to which perceivers used those cues as signs ofdeceit. When senders pretended to like people they really disliked, theirdescriptions were less positive and more neutral than when they honestlydescribed people they really did like. When feigning disliking, senders utteredmore nonfluences than when expressing honest disliking. All of these cueswere used by perceivers in their judgments of deceptiveness; in addition,perceivers judged as deceptive descriptions that were spoken slowly and con-tained many Um's and er's. Expressions of liking that contained many other-references, few self-references, and many nonspecific (undifferentiating)descriptors were also perceived to be deceptive. To facilitate the study of ac-tual and perceived cues deception, and their correspondence, a heuristic modelwas proposed.

We thank Russell A. Eisenstat, Susan Finkelstein, Michael Fridkin, and Donald B. Rubinfor their help with this research. Requests for reprints should be sent to Bella M. DePaulo,Department of Psychology, Gilmer Hall, University of Virginia, Charlottesville, Virginia22901.

292 DBPAULO, ROSENTHAL, ROSENKRANTZ, GREEN

In a recent review of cues to deception. Kraut (1980) has pointed out thatmost cues that have been studied have been nonverbal ones; verbal cues, bycontrast, are noticeably underrepresented in the extant literature. In fact.Kraut found only one verbal cue (amount of detail) that had been examinedin more than one study, and that cue did not reliably distinguish truth fromdeception across studies. There are probably several reasons for this em-phasis on nonverbal rather than verbal cues. First, there has been anassumption in much of the research on deception that nonverbal cues, sincethey are presumably less readily controlled by the liar, might be morerevealing of the liar's true affective state than are verbal cues. Historically,this notion has been traced back to Darwin (1872) and to Freud (1959);more recently, it has been elaborated in the theory and research of Ekmanand Friesen (1969; Ekman, 1981). A second reason why verbal cues mightbe underrepresented in research on cues to deception is a more practicalone; that is, the scarcity of widely used and theoretically relevant scoringsystems for measuring and categorizing verbal content.

The accumulating evidence on the perpetration and detection of decep-tion suggests both that the verbal stream might in fact contain importantcues to deceit and that perceivers do use verbal cues in their judgmentsabout other people's truthfulness. DePaulo, Zuckerman, and Rosenthal(1980) recently reviewed more than 20 studies of perceivers' accuracy atdetecting lies from either face cues, body cues, tone of voice (filteredspeech), verbal cues (words only), audio cues (words plus tone of voicecues), or audiovisual cues (words plus tone plus visual cues). They conclud-ed that perceivers were strikingly more accurate at detecting lies when theyhad access to verbal cues (verbal, audio, and audiovisual conditions) thanwhen only nonverbal cues were available (face, body, and tone conditions).Of the three modalities that included words, accuracy was highest in theaudio condition (although not significantly higher than in the verbsil oraudiovisual conditions). These results suggest that the speech of liars mightcontain cues that can be effectively used by perceivers in their attempts todetect lies. Corroborating evidence comes from a study in which allperceivers were given full audiovisual access to the liars but some of theperceivers were told to pay special attention to the visual cues, others weretold to pay special attention to the words, others were told to pay special at-tention to the tone, and the rest were given no special attention instructions(DePaulo, Lassiter & Stone, 1982); in that study, perceivers were more ac-curate at detecting lies in the attend-to-words and (especially) in the attend-to-tone conditions than in the visual or control conditions. Thus, perceiversare especially accurate at detecting hes when verbal cues are available tothem (DePaulo et al., 1980) and they can improve somewhat on their usualaccuracy when advised to pay special attention to some aspect of the streamof speech (DePaulo et al., 1982).

ACTUAL AND PERCEIVED CUES TO DECEPTION 293

Ekman, Friesen, O'SuUivan, and Seherer (1980) studied the relative im-portance of face, body, and speech (words plus tone) cues in judgments ofthe personality and affect of people who were either lying or telling thetruth. They found, for three samples of perceivers, that when the senderswere lying, judgments made from speech cues correlated more highly withjudgments made from the "whole person" (audiovisual condition) than didjudgments made from either face or body cues. This suggests that when for-ming impressions of people who are lying, perceivers may be more influenc-ed by the liars' speech than by their face or body cues.'

The findings that speech cues do reliably reveal whether or not deceptionis occuring and that perceivers are strongly influenced by speech cues intheir judgments of liars suggest three further questions:

1. What specific verbal or paralinguistic cues actually do distinguishdeception from truth;

2. What verbal or paralinguistic cues do perceivers use in their judgmentsabout honesty and dishonesty; and, most importantly,

3. What is the correspondence between the cues that really do distinguishdeception from truth ("actual cues to deception") and the cues thatperceivers use in their judgments ("perceived cues to deception")? Examin-ing the role of specific cues as both actual and perceived indicators of decep-tion will yield a more fine-grained portrayal of the ways in which deceptionis perpetrated, diagnosed, and misdiagnosed.

Although numerous studies have examined cues that might actuallydistinguish truth from deception, such as response latency, response length,rate, pitch, and fluency, and several other studies have investigated many ofthese same cues as they are used by perceivers in their judgments aboutdeceit, there are fewer instances in which actual cues to deception andperceived cues to deception have been examined in the seime study. Even inthese instances, very few of the cues were verbal cues (see Knapp & Com-adena, 1979; Kraut, 1980; and Zuckerman, DePaulo & Rosenthal, 1981, forreviews). Of the studies that have examined both actual and perceived cuesto deception, the three that examined more than two cues (Kraut, 1978;Kraut & Poe, 1980; Krauss, Geller & Olson, Note 1) have all shown that thenumber of cues that perceivers used to judge deception (to a statisticallysignificant degree) exceeded the number of cues that did reliably distinguishtruth from deceit.

We would like to suggest the use of two modes of presentation andanalysis which might facilitate the study of cues to deception. First, an exactquantitative assessment, for each cue, of the degree of association between

'A similar conclusion has been reached based on studies not involving deception. Thus,Krauss, Apple, Morency, Wenzel, and Winton (1981) have shown, in two studies, thatperceivers' evaluative judgments are better predicted by verbal than by visual information.

294 DEPAULO, ROSENTHAL, ROSENKRANTZ, GREEN

the cue and deception, and between the cue and perceived deceit, should beprovided. These two sets of numbers, representing the extent to which eachcue is associated with deception, and is perceived to be associated withdeception, can then be correlated with each other to yield a single correla-tion coefficient indicating the degree of correspondence between actual andperceived cues in that study. This correlation will of course vary with theparticular cues that the researchers have decided to examine, but it does ac-curately and succinctly summarize the covariance between actual andperceived cues for the particular set of cues in question.

Actual cues to deception can be conceptualized as falling into threecategories:

1. deceptors: behaviors more likely to occur during deceptive responsesthan during truthful ones;

2. neutrals: behaviors neither more nor less likely to occur during decep-tive responses than during truthful ones; and

3. Sinceritors: behaviors less likely to occur during deceptive responsesthan during truthful ones.

Similarly, perceived cues to deception can also be conceptualized as fall-ing into one of three categories:

1. discreditors: behaviors more likely to be interpreted as cues to decep-tion than as cues to truth;

2. neurtrals: behaviors neither more nor less likely to be interpreted ascues to deception than as cues to truth; and

3. creditors: behaviors less likely to be interpreted as cues to deceptionthan as cues to truth.

Using this conceptual scheme, each cue of interest can be located in a 3 x 3table which summarizes, for each cue, whether it is in fact a deceptor, asinceritor, or a neutral cue, and whether it functions for perceivers as adiscreditor, a creditor, or a neutral cue.

In the present study we examined actual and perceived verbal and paral-inguistic cues to deception among 40 people (20 males and 20 females) whoserved both as liars (senders) and perceivers. As senders, subjects describedeither honestly or dishonestly other people that they knew. So that decep-tion would vary independently of the particular affect being conveyed,senders described people they liked and people they disliked (honest descrip-tions), and they also described the person they liked, pretending to dislikehim or her, and the person they disliked, pretending to like him or her(deceptive descriptions).

In choosing the particulau" cues to study, we were guided by considera-tions of the kinds of internal states that might covary with deception. For


example, if it is cognitively more challenging to tell a lie than to tell thetruth, then liars might experience heightened cognitive processing. Liarsmight also experience anxiety or guilt during the proces of deceiving (seeKraut, 1980). Cues that might be associated with anxiety were fairly easy togenerate. Previous research has shown that speech disturbances and variousrate-related variables are sensitive to variations in anxiety (e.g., Goldman-Eisler, 1961a; Kasl & Mahl, 1965; Mahl, 1956); thus, in the present study,we measured nonfluences, um's and er's, rate, and rate change. It shouldbe noted, however, that these kinds of behavior might also be affected bycognitive processing demands (see, for example, Goldman-Eisler, 1961b),or perhaps even by guilt.

If people experience guilt during deception, they might deal with it by be-ing noncommittal and dissociating themselves from their responses,perhaps by making more neutral (rather than evaluative or extreme)statements, or by speaking in the third person, or by describing people (inthe dishonest conditions) in global and undifferentiating ways. To measurethe evaluativeness, personal involvement, and degree of differentiation inthe deceptive and nondeceptive descriptions, we used a coding systemdeveloped and validated specifically for use in studying free verbal descrip-tions (Peevers & Second, 1973). We also developed a measure of extremestatements (e.g., use of superlatives).

The degree of differentiation in person descriptions has been shown to in-crease systematically with age; thus, this measure, too, might be sensitive tovariations in cognitive demands. Finally, we also investigated the possibilitythat the use of the words "but" and "yet" in descriptions of others mightalso indicate a more cognitively complex mode of construing others, andhence might also vary with the cognitive demands of the task.

METHODS

Subjects

Subjects were 40 undergraduate summer school students (20 males and 20females) recruited for a study of "person descriptions" and paid for theirparticipation.

Procedure

Subjects were asked to take 1 minute to describe each of the following per-sons: someone they liked, someone they disliked, someone they felt am-bivalent about, and someone they felt indifferent about. (Ambivalence wasdefined as strong feelings of both liking and disliking; indifference was


defined as no strong feelings of liking or disliking.) To elicit deception, sub-jects were also asked to describe the person they liked as if they really dislik-ed him or her (pretend to dislike condition) and to describe the person theydisliked as if they really liked him or her (pretend to Uke condition). Thesesix descriptions were given in one of nine different orders (randomly assign-ed). Half of the subjects (10 males and 10 females, randomly assigned)described males; the others described females.

During the sessions, the experimenter remained behind a one-way mirrorand videotaped the descriptions. Subjects were aware of this, and also knewthat the experimenter was not informed of the sequence of their descrip-tions. They were urged to try to be very convincing in all of their descrip-tions.

From these descriptions, two one-hour videotapes were made. Each tapeincluded all six descriptions from 20 of the senders (10 males and 10females). Thus, each tape consisted of 120 descriptions. The order of ap-pearance of the senders on these tapes were randomized. The middle 20seconds of each description,^ plus rating pauses, were included on the tapes.All subjects returned to judge one of these videotapes. Subjects alwaysjudged a videotape on which they did not appear. Subjects rated each seg-ment on a 9-point scale of deception, with endpoints labeled "speaker is notvery deceptive" (1) and "speaker is very deceptive" (9). (Subjects rated thedescriptions on other scales, too, but those were not relevant to the presentstudy.)^

The search for verbal and paralinguistic cues to deception using theseperceivers and these stimulus materials is particularly appropriate forseveral reasons. First, previous research (DePaulo & Rosenthal, 1979) hasshown that these perceivers performed substantially better than chance atdetecting each other's deception from audiovisual cues (which includespeech cues). Second, a subsequent investigation (DePaulo, Rosenthal,Green & Rosenkrantz, 1982) involving different subjects who judged these

^Only the middle 20 seconds were used to keep the rating task to a more manageable length.The available evidence suggests that this was not an unreasonably short segment. First, theaverage number of words spoken in these 20-second segments was 37.51. And second,perceivers showed substantial accuracy at detecting deception from these segments (DePaulo &Rosenthal, 1979).

'Subjects juding the tapes did know that two of the six descriptions made by each speakerwould be deceptive. In this way, their task may have been somewhat simpler than the taskfaced by people trying to detect lies outside of the laboratory, where the base rates for the occur-rence of deception are usually completely unknown. However, since the messages that subjectswere asked to judge included several different types (ambivalent messags, indifferent messages,unambivalent truths, lies of liking and lies of disliking), subjects' task was more difficult (andmore reahstic) than that used in the typical deception paradigm in which subjects are askedmerely to decide whether the speaker is lying or telling the truth.


descriptions from access to either face cues, tone of voice cues, verbalcues, audio cues, or audiovisual cues showed that, consistent with pastresearch (see DePaulo et al., 1980), deception was most readily detectedfrom any source of cues that included words (i.e., the verbal, audiovisual,and particularly the audio, conditions). Thus, there was reason to believethat there would be important verbal and paralinguistic cues to deception inthese particular stimulus materials.

Verbal and Paralinguistic Cues

The verbal and paralinguistic cues measured in the present study are definedbelow.

1. Nonfluencies. Number of sentence changes, superfluous repeti-tions, stutterings, sentence incompletions, and other speech disturbances, asdescribed and defined by Kasl and Mahl (1965). The number of nonfluen-cies in each description was divided by the total number of words in eachdescription.

2. Um's and er's. Total number of um's, uh's, ah's, and er's, dividedby the total number of words.

3. Rate. Since each excerpt that subjects judged was 20 seconds long,rate was readily assessed by simply counting the number of words per seg-ment.

4. Number of sentences. The number of complete sentences in eachsegment. This measure would be useful if perceivers were sensitive not somuch to the number of words per unit of time, but to a related higher-ordervariable, such as number of sentences. (Or, it would be useful if deceptiveresponses differed from truthful ones in this way.)

5. Rate change. The number of words in the second half of thedescription minus the number of words in the first half of the description.The rationale is that if senders telling lies start out having a difficult timeconstructing their lies, they might try to cover over the resultinglaboriousness by quickening their rate of speech toward the end of theirdescription.

The following 13 categories were borrowed or derived from the PersonConcept Code (Peevers & Secord, 1973). The descriptions of each categoryare adapted from Peevers and Secord (1973, p. 122). The first four variablesare levels of "descriptiveness." Descriptiveness refers to the amount of in-formation given about a person as a unique individual. The number of


items at each level for each description was divided by the total number ofdescriptive items for that description.

6. Undifferentiating. The person was not differentiated from his orher environment, but was described in terms of possessions or social setting.(Examples: "Mary lives in a big house" and "John has a nice mother.")

7. Simple differentiating. The person was differentiated as an in-dividual, but was described in terms of simple, superficial characteristics,global judgments, or his or her relationship to the perceiver. (Examples:"Mary is nice" and "John likes me.")

8. Differentiating. The person was described in terms of fairlyspecific personal characteristics, such as interests, abilities, beliefs, or tem-porary states or conditions. (Examples: "Mary is a good athlete" and"John is confused right now.")

9. Dispositional. The person was described in terms of traits whichhad implications for his or her behavior in a wide range of situations. (Ex-ample: "John is talkative.")

10. Cognitive complexity. This measure summarized the descrip-tiveness across the four different levels, weighing the cognitively complexlevels more heavily than the cognitively simpler levels. More specifically, theformula was completed as follows: cognitive complexity = undifferen-tiating + (2 X simple differentiating) + (3 x differentiating) + (4 x disposi-tional). This measure would be useful if perceivers were sensitive primarilyto the overall level of sophistication or differentiation in the descriptions (orif deceptive responses differed from truthful ones in this way).

The following three items describe the senders' degree of personal in-volvement in their descriptions. Personal involvement, according to Peeversand Secord (1973), refers to the frame of reference that individuals adopt indescribing others; that is, the degree to which they involve themselves inwhat they say about others. The number of items at each level for eachdescription was divided by the total number of descriptive items for thatdescription.

11. Egocentric. The other person was described in subjective, self-oriented terms. Most descriptions in which " I " or "me" were used were ofthis type. (Example: "She gave me a cookie.")

12. Mutual. Another person was described in terms of his or her rela-tionship to the perceiver. Descriptions of this type were often characterizedby the use of "we" or "us." (Example: "We go bike-riding together.")


13. Other-oriented. No personal involvement was expressed by theperceiver. (Example: "He has blue eyes.")

14. Excess other. The degree to which the sender gave more other-oriented than egocentric characterizations, as defined by a difference score(other-oriented minus egocentric).

Items 15, 16, 17, and 18 assess the evaluative tenor of the descriptions.Each item of information in each description is coded as either positive,negative, or neutral according to whether the item describes desirable,undesirable, or evaluatively neutral characteristics. Details of the scoringare provided in the Peevers and Secord Person Concepts Code (see Peevers& Secord, 1973).

15. Positive. Number of positive items, divided by the total numberof descriptive items.

16. Negative. Number of negative items, divided by the total numberof descriptive items.

17. Neutrai. Number of evaluatively neutral items, divided by thetotal number of descriptive items.

18. Excess positive. The degree to which senders said more positivethan negative things in their descriptions, as defined by a difference score(positive minus negative).

19. Extremes. The person is described as extreme on a certain dimen-sion. (Examples: She is extremely aggressive. He is very very cold. She isthe most intelligent person I know.) Number of extremes was divided by thetotal number of words.

20. But's and yet's. Number of occurrences of the word "but" or"yet," divided by the total number of words.

For each cue, all descriptions were scored by a single judge. In order toassess the reliability of each measure, a second judge was recruited to scorea subset of the descriptions (the six descriptions of six of the senders). Theratings made by both judges were always based on minimal cues—that is,exact transcripts. Product-moment correlations between the two judges'ratings were computed separately for each of the six types of descriptions.The interjudge reliabilities were: nonfluencies, .89; um's and er's, 1.00;rate, 1.00; number of sentences, .98; descriptiveness, .90; personal involve-ment, .97; evaluative tenor, .66; extremes, .98; and buts and yets, 1.00. Onthe whole, then, interjudge reliability seemed quite satisfactory.


RESULTS

Descriptions were the units of analysis for the computation of actual andperceived cues to deception. The descriptions analyzed were those in whichthe senders were telling the truth (the like and dislike descriptions) or lying(the "pretend to like" and "pretend to dislike" descriptions). Thus, alltogether there were 160 relevant descriptions (40 senders x 4 descriptions persender). In all of our analyses, we examined separately descriptions in whichthe expressed affect was Hking (i.e., the honest liking descriptions and the"pretend to like" descriptions) and those in which the expressed affect wasdisliking (i.e., the honest disliking descriptions and the "pretend to dislike"descriptions).

For each description, the following information was available: (1) therate of occurrence of each cue; (2) whether the description actually washonest or deceptive; (3) whether the affect that the sender was attempting tosend—that is, the expressed affect—was liking or disliking; (4) the sex of thesender; (5) the mean deception rating made by all judges; (6) the meandeception rating made by the male judges; and (7) the mean deceptionrating made by the female judges.

Actual Cues to Deception

Columns 1 and 3 of Table 1 show, for each cue, the correlation between (1)the rate of occurrence of that cue in each description, and (2) whether thedescription was honest or deceptive. In column 1, a positive correlation in-dicates that the cue occurred more often in descriptions in which senderswre pretending to like people they actually dishked than in honest descrip-tions of people the senders really did like. Analogously, a positive correla-tion in column 3 indicates that the cue occurred more often in "pretend todislike" descriptions than in honest dislike descriptions. Thus, positive cor-relations always indicate that the cue occurred more frequently in the decep-tive descriptions than in the truthful ones.

Column 1 shows that when senders were pretending to like people theyreally disliked, they made more affectively neutral comments and fewerpositive comments than when they were honestly describing people theyreally did like. Also, the degree to which the number of positive commentsexceeded the number of negative comments was not as great when thesenders were lying as when they were telling the truth.

Column 3 shows that only one cue distinguished truth from deceptionwhen the expressed affect was disliking. Senders produced more speechnonfluencies when they were pretending to dislike people they really did likethan when they honestly described people they really did dislike.

To compare the cues associated with lies of liking to the cues associatedwith lies of disliking, the correlations in column 1 were correlated with thecorrelations in column 3. (In this analysis, and in all similar analyses, the


TABLE 1Verbal and Paralinguistic Correlates of Actual

Deception and of Perceived Deception

Verbal or Paralinguistic Cue

1. Nonfluencies

2. Um's and er's

3. Rate

4. Number of sentences

5. Rate change

6. Undifferentiating

7. Simple differentiating

8. Differentiating

9. Dispositional

10. Cognitive complexity

11. Egocentric

12. Other

13. Mutual

14. Excess other

15. Positive

16. Negative

17. Neutral

18. Excess positive

19. Extremes

20. But's and yet's

ActualDeception

.08

.09

.02

-.02

.08

.08

-.12

.13

-.02

.03

-.16

.20

-.12

.17

- . 3 0 «

-.01

.31*

-.28}

-.14

.00

Expressed Affect

Like

PerceivedDeception

.19

.46**

-.56***

-.21

-.03

.11%

.01

.05

-.15

-.13

-.29}}

.28}

-.05

.29}}

- .31*

-.03

.32*

-.29}}

.03

-.05

Dislike

ActualDeception

.23}

.07

- .09

-.05

.05

.02

.07

-.01

-.07

-.08

.08

-.13

.11

- .10

.16

-.04

-.04

.09

-.02

-.11

PerceivedDeception

.24}

.25}

-.33*

-.07

-.03

.09

.11

-.03

-.09

-.13

-.00

-.08

.19

-.04

.13

-.20

.16

.20

.20

- .02

Note: Using descriptions as the units of analysis, A = 80 (there is some variation due tomissing data). Significance levels based on these Afs are indicated by daggers. However, sinceeach sender contributed two descriptions to each correlation, a more conservative estimatewould involve basing the dfon the number of senders. Significance levels based on this iV (ap-proximately 40) are indicated by asterisks.tp < .05, df =78 *p < .05, df s 38

}}/?< .01,rf/=78 **p< .Ol,rf/=38•**p< .001, d / = 38

correlations were first transformed by Fisher's Z.) The resulting correlationwas negative and significant, r (18) = - .46, p < .05. Thus, the cues thatindicate that a person is feigning liking are not the same cues that indicatethat a person is feigning disliking.


To exEimine sex differences in the ways that males and females lie, actualcues to deception were computed separately for male and female senders.The results are shown in Table 2. When expressing liking, males andfemales showed a marginally significant degree of similarity in the ways thatthey lied, r (15) = .42, p - .09 (correlation of column 1 with column 2).That is, the degree to which particular cues occurred more frequently whensenders were pretending to like people (compared to when they really didlike them) was somewhat similar for male and female senders. However,when the expressed affect was disliking, the comparable correlation wasnonsignificant and slightly negative, r (15) = - .25 between columns 3 and4. Thus, the degree to which these cues distinguished honest disliking fromfeigned disliking was not at all similar for male and female senders.

A closer examination of Table 2 indicates the specific ways in whichmales and females deceive differently. When expressing liking, women(compared to men) say fewer positive things—in both in an absolute senseand in comparison to negative things—when they are lying than when theyare telling the truth. (Z for the difference between males and females =2.48, p = .01, for positive, and 2.53, p = .01 for excess positive. Analyseswere based on the twp messages—truthful and deceptive—sent by each ofthe 20 male or female encoders). Similarly, when expressing disliking,females say fewer negative things when lying than when telling the truth (Z= 2.01, p = .04). Thus, when deceiving, females make less evaluatively ex-treme comments than they do when telling the truth. Women, compared tomen, also say more neutral things when they are lying than when they aretelling the truth, and this is true for both types of lies (Z = 2.48, p = .01 forexpressed liking, and Z = 2.69, p = .008, for expressed disliking).

To test the overall power of this particular set of verbal and paralinguisticcues in distinguishing truth from deception, simultaneous regressions werecomputed. The predictor variables were the frequencies of the cues in eachdescription" and the dependent variable was the actual deceptiveness of eachdescription (i.e., either truthful or deceptive). This regression was com-puted separately for male and female senders and for expressed liking and

'The 20 cues were factor analyzed and z-scored before being entered into the regressionequation. A nine-factor solution provided the best fit. The factors were: (1) nonfluencies, um'sand er's, rate (negatively loaded), and number of sentences (negatively loaded); (2) other, ex-cess other, and egocentric (negatively loaded); (3) dispositional, cognitive complexity, and sim-ple differentiated (negatively loaded); (4) positive, excess positive, and negative (negativelyloaded); (5) extremes, but's and yet's, and rate change; (6) neutral; (7) undifferentiated; (8) dif-ferentiated; and (9) mutual. When a factor was defined by more than one variable, the mean ofthose variables (reversed when necessary) was entered into the regression equations. Df mayvary slightly across analyses because of missing data.


TABLE 2Sex Differences in Verbai and Paralinguistic

Correiates of Actual Deception

Verbal or Paralinguistic Cue

1. Nonfluencies

2. Um's and er's

3. Rate

4. Number of sentences

5. Rate change

6. Undifferentiating

7. Simple differentiating

8. Differentiating

9. Dispositional

10. Cognitive complexity

11. Egocentric

12. Other

13. Mutual

14. Excess other

15. Positive

16. Negative

17. Neutral

18. Excess positive

19. Extremes

20. But's and yet's

Males

.09

-.00

.07

.06

.00

- .09

-.08

.16

-.05

.02

-.10

.20

-.18

.14

-.06

-.02

.07

-.05

-.20

.00

Expressed Affect

Like

Females

.06

.18

-.05

-.10

.18

.26

-.15

.10

.00

.04

- .21

.19

.20

-.57***

.58***

-.57***

-.06

Males

.28

.10

-.03

.00

-.12

.02

-.09

.11

-.02

.03

.09

-.18

.16

-.14

.19

.18

-.36*

-.05

-.09

-.16

Dislike

Females

.18

.05

-.18

-.10

.26

.24

-.14

-.13

-.22

.06

-.06

-.05

.16

-.28

.25

.28

.07

Note: A null entry indicates that the correlation could not be computed. For example,females expressing liking never said anything negative, whether lying or telling the truth.

expressed disliking. In these regressions, then, the 40 units of analysis werethe two messages (truthful and deceptive) sent by each of the 20 male orfemale encoders. Results showed that these cues significantly differentiatedtruth from deception only for females expressing Hking, F(8,28) = 2.71, p= .02, R = .44. (For females expressing liking, R = .23; for males ex-pressing liking and disliking, R = .16 and .32 respectively).


Perceived Cues to Deception

Columns 2 and 4 of Table 1 show, for each cue, the correlation between (1)the rate of occurrence of that cue in each description, and (2) the meanrating of deceptiveness of each description (i.e., perceived deceptiveness).

Column 2 shows that when senders were describing people they like orpeople they were pretending to like, they were perceived as especially decep-tive when they used more um's and er's, talked more slowly, and describedothers in more undifferentiating terms. Infrequent use of egocentricreferences (both in an absolute sense and in comparison to "other"references) and frequent use of "other" references also made the descrip-tions appear more deceptive to the perceivers. Finally, less evaluatively ex-treme descriptions were also perceived as more deceptive, as evidenced bythe negative correlations with "positive" and "excess positive" and thepositive correlation with "neutral." In summary, when senders are en-coding positive affect, they are seen as more deceptive when their descrip-tions are characterized by few positive and many neutral terms; feweregocentric references and more other references; more undifferentiatingterms, more um's and er's, and a slower rate.

Column 4 shows that three cues were related to perceptions of decep-tiveness when the expressed affect was disliking. Consistent with the resultsfor the "expressed liking" descriptions, descriptions were perceived as moredeceptive when they were uttered more slowly and when they containedmany um's and er's. For expressions of disliking, the occurrence ofnonfluencies was also directly related to perceptions of deceptiveness.

To compare the cues used in perceptions of lies of liking to the cues usedin perceptions of lies of disliking, the correlations in column 2 were cor-related with the correlations in column 4. The resulting correlation waspositive and significant, r (18) = .46, p < .05. Thus, perceivers use cuesvery similarly in making judgments about different kinds of lies.

To examine sex differences in the ways that males and females perceivedeception, columns 2 and 4 (of Table 1) were computed separately for maleand female perceivers. For expressed liking, the correlation between the 20perceived deception correlations based on the male perceivers' judgmentsand corresponding correlations based on the female judgments was .85 {p< .001). For expressed disliking, the correlation was .52 (p < .001). Thus,males and females showed substantial agreement in the way that they usedthis set of cues in their judgments of deceptiveness.

For expressions of liking, the verbal and paralinguistic cues examined inthis study accounted for a significant portion of the variance in judges'ratings of deceptiveness. This was true across all combinations of sender sexand perceiver sex. (For males and females judging male senders, R = .48and .50, respectively; for males and females judging female senders, R =


.62 and .58, respectively. F's [8,28] ranged from 3.07 to 5.65, all/j's < .01).For expressions of disliking, these cues did not account for a significantportion of the variance in perceived deceptiveness. (R^ = .19, .35, .15, and.31, respectively, for males and females judging males, and males andfemales judging females.) The weak results for expressions of disliking, ob-tained in the analyses of both actual and perceived cues, may be a functionof the relative infrequency of this type of deception. It seems plausible thatpeople more often pretend to like disliked others than the reverse, althoughthis has yet to be demonstrated empirically.

The Relationship Between Actuai and PerceivedCues to Deception

Columns 1 and 3 of Table 1 indicate the degree to which each cue is actuallyassociated with deception, while columns 2 and 4 indicate the degree towhich each cue is perceived to be associated with deception. A precise,quantitative asssessment of the correspondence between actual and perceiv-ed cues can be obtained by correlating the "actual" column with the"perceived" column. For both expressed liking and expressed disliking, thecorrespondence was substantial, r (18) = .66, p < .001 for expressed likingand r (18) = .69, p < .001, for expressed disliking. Thus, the extent towhich each of these various cues actually did distinguish between truth anddeception was strongly related to the extent to which they were perceived asdoing so.'

Table 3 shows the relationship between actual and perceived cues todeception separately for male and female perceivers judging male andfemale senders. Overall, the correspondence between actual and perceivedcues was very similar for female and male perceivers, for same sex and op-posite sex pairs, and for expressed liking and expressed disliking. Althoughthe correspondence did tend to be higher for female senders than for malesenders, this difference, too, failed to reach significance, Z = 1.27, p =.20, two-tailed.

'These correlations may appear to be inflated for several reasons. First, the list of cues in-cludes three variables that were derived from other variables in the list—that is, cognitive com-plexity, excess other, and excess positive. However, when these derived variables were omitted,the correlations changed very little: r (15) for expressed liking = .62, p = .008; for expresseddisliking, r (15) = .65, p = .004. Also, when the nine factors (see footnote 4) were used asunits, the relationships were still positive and highly significant, r (7) for expressed liking =.71, p = .03; for expressed disliking,/• (7) = .87,/) = .002. Second, the correlations may seemto result from the fairly large number of variables that were not substantially associated witheither actual deception or perceived deception. However, even when the actual and perceivedcorrelations were small, they were usually in the same direction. For expressed liking, 14 ofthe20 cues were correlated with actual and perceived deception in the same direction; for express-ed dishking, 15 of the 20 cues shared the same signs.


TABLE 3Reiationship Between Actual Cues to Deception

and Perceived Cues to Deception

Sex of sender

Sex of perceiver

Male

Female

Male

.34

.42 .

Like

Expressed Affect

Female

.61**

.86***

Male

.45*

.33

Dislike

Female

.59*

.66**

*p < .05**p < .01

***p < .001

DISCUSSION

Table 4 provides both a summary of our results and a heuristic model forconceptualizing simulateneously both actual and perceived cues to decep-tion. The columns of Table 4 represent the three types of actual cues todeception (deceptors, neutrals, and sinceritors), while the rows represent thethree types of perceived cues to deception (discreditors, neutrals, andcreditors).

Some of the variables that we included largely on the basis of theirhypothesized or demonstrated association with anxiety—that is, nonfluen-cies, um's and er's, and rate—were among the most powerful of our predic-tor variables. In expressions of disliking, speech nonfiuencies served both asdeceptors and discreditors. That is, they were interpreted as signs of decep-tion and they actually were signs of deceptions. Um's and er's were directlyassociated with perceptions of deception both when liking and dislikingwere the expressed affects. These results are consistent with those reportedin recent reviews of the literature (Kraut, 1980; Zuckerman et al., 1981).Thus, the hemming and hawing that our culture tends to associate withequivocation and deception does indeed appear to figure prominently inpeople's judgments about deception, and it also appears to characterizesome of the kinds of lies that have been studied by psychologists.

Rate of speaking was not an actual cue to deception in this study, but itdid serve as a creditor to perceivers, who rated fast talkers as significantlyless likely to be lying than slow talkers. In research not dealing directly withdeception, fast talkers have also been imbued with an array of desirable at-tributes; in a study of Miller, Maruyama, Beaber, and Valone (1976),speakers talking more quickly were perceived as more objective, intelligent,and knowledgeable. In two studies using computerized manipulations of

ACTUAL AND PERCEIVED CUES TO DECEPTION 3 0 7

TABLE 4Heuristic iVIodei of Actuai and Perceived

Cues to Deception

Perceived Cuesto Deception

Discreditors

Neutrals

Creditors

Deceptors

Nonfluencies^

Neutral*"

Actual Cues to DeceptionNeutrals

Um's and Er'sUndifferentiating"'Other''Excess other*"

Number of SentencesRate ChangeSimple Differentiating,

Differentiating,Dispositional, andCognitive Complexity

MutualNegativeExtremesButs and yets

RateEgocentric''

Sinceritors

Positive''Excess Positive''

''This categorization describes the results for expressed disliking. For expressed liking, thiscue was neutral with respect to both actual and perceived deception.

''This categorization describes the results for expressed liking. For expressed disliking, thiscue was neutral with respect to both actual and perceived deception.

speech rate, curvilinear relationships emerged, such that deviations in eitherdirection from the normal unmanipulated rate were judged as lessbenevolent (Smith, Brown, Strong, & Rencher, 1975), less persuasive, andmore deceptive (Apple, Streeter, & Krauss, 1979). Perhaps perceivers useany plausible signs of anxiety—for example, speech disturbances, um's ander's, or unnaturally slow (or fast) rate—as cues to deception. Consistentwith this interpretation, there is evidence that variations in rate affect at-tributions of nervousness in the same way that they affect attributions ofdeceit—that is, unusually slow (or fast) talkers are perceived as more decep-tive and as more nervous (Apple et al., 1979)—and that speakers judged tobe particularly nervous are also judged as especially likely to be lying (Kraut& Poe, 1980; Krauss et al.. Note 1).

In the present study, we investigated the possibility that some of the cuesmight be curvilinearly associated with actual or perceived deception.' Inmost cases, these curvilinear correlations either were not significantly

'Curvilinearity was coded by j-scoring each cue separately, then taking the absolute value ofthe z-scores.


associated with actual or perceived deception, or they were significantlyassociated with the linear correlations (as in the case of extremely skeweddistributions). The rate variable, however, was a notable exception. For ex-pressed liking, the curvilinear component was substantially associated bothwith actual deception,/• (78) = .21, p = .06, and with perceived deception,/•(78) = .25, p = .03. (Descriptions were the units of analysis.) Moreover,the curvilinear and linear components of the rate variables were indepen-dent, r (78) = - .06. Thus, people expressing positive affect who are speak-ing too quickly or too slowly are likely to be lying, and they are also Ukely tobe perceived as lying.

Although speech errors and variations in rate may be actual or perceivedindicants of anxiety, they might also be linked to cognitive processing or toguilt. Other variables that have emerged in this study as discreditors,creditors, deceptors, or sinceritors strengthen the case for the role of anxietyor guilt as predictors of valid cues to deception. These variables—that is,the use of less evaluatively extreme descriptions, more neutral descriptions,fewer self-references, more other references, and more undifferentiateddescriptive terms—all seem to be verbal devices that senders can use todissociate themselves from the content of their messages. Senders feigningliking, for example, were not quite as positive in their descriptions as theywere when expressing genuine liking; they also used more neutral items.Perceivers seemd to pick up on this: They rated these more noncommittalmessages as more deceptive. When senders described others in verynonspecific, undifferentiating ways—for example, when they describedtheir clothes, possessions, or jobs rather than their interests, abilities, ortraits; and when they did not involve themselves personally in their descrip-tions (that is, when they used the third person rather than the first)—theywere also perceived as being deceptive, even though these behaviors werenot valid clues to deception.

Cues that signal dissocation, distancing, evasiveness, or a lack of involve-ment or commitment seem to play an important role in the perpetration (seealso Kuiken, 1981) and detection of deceit. These kinds of cues could resultfrom senders' guilt about lying, or, as Kuiken (1981) has suggested, from amore affectively neutral "apprehensive[ness] about affirming the validity ofthe contents of [one's] own statements" (p. 183). Finally, these verbaldistancing devices, as well as the speech errors and rate fluctuations, mightbe indicative of the senders' insecurity about their ability to lie successfully.Although this suggestion is merely speculative, Kuiken's data (1981) are con-sistent with it: In a study in which self-esteem was experimentallymanipulated, the low self-regard subjects used more "nonimmediate"(distancing) speech than the high self-regard subjects.

As in earlier studies (e.g.. Kraut, 1978; Kraut & Poe, 1980), the percep-tion of deception was much more powerfully predicted than was actual


deception. Kraut (1980), arguing from an evolutionary standpoint, has sug-gested that there may be no reliable cues to deception, since it would bedisadvantageous for senders to have their deception attempts readilydetected. While we certainly agree that no cue is likely to signal deceptionacross all senders, perceivers, social settings, and types of lies (cf. DePaulo& Rosenthal, 1979), within any given situation in which perceivers are ableto detect deception successfully, it seems reasonable to suppose that reliablecues to deception could be found. These cues might vary systematicdly withaspects of the deception situation. Perhaps the cues would be associatedwith deception in more complex ways than those we have studied in the pre-sent investigation.

Perceptions of deception might also be governed by intricate and interac-tive rules: For example, a certain behavior might influence judgments ofdeceptiveness only for certain kinds of perceivers and only when emitted bycertain kinds of liars in certain kinds of contexts. However, research onhuman inference and decision making in other contexts, suggests thatjudgmental processes often follow rather simple and straightforward rules(e.g., Slovic & Lichtenstein, 1971). Perhaps perceived cues to deceit couldbe most effectively identified by a knowledge of people's lay theories aboutdeception. If, for example, perceivers believe that people who are lying arenervous and evasive, then their impressions of nervousness and evasivenessmight be the most powerful predictions of their judgments about deception.The cues to perceived deceit, then, would be whatever cues give perceiversthe impression that the speaker is nervous and evasive. Further, as Hockingand Leathers (1980) have suggested, the cues that are stereotypicallyassociated with deceit may be the cues that deceivers try most carefully tomonitor and control.

The present investigation, as well as others that have preceded it (e.g..Kraut & Poe, 1980), has documented considerable consistency in the use ofcues by different kinds of perceivers. For example. Kraut & Poe (1980), intheir study of deception detection among customs officials and laypersons,have found that very similar "search" rules are used by the officials and thelaypersons, by the more and the less experienced officials, and by high andlow self-monitors. In the present study there was substantial agreement be-tween male and female perceivers in the cues they used in their judgmentsabout deceit. Despite this similarity in the use of cues, previous research(DePaulo & Rosenthal, 1979; Littlepage & Pineault, 1979) has established anear-zero correlation between accuracy of detecting women's lies and ac-curacy at detecting men's. These findings led DePaulo and Rosenthal (1979)to speculate that perceivers may operate on the assumption that a single setof cues is consistently associated with deception, when in fact women'sdeceit is revealed in different ways than men's. The present resultsdemonstrate that, at least in some ways, women do in fact lie differently


than men. Specifically, when deceiving, women, more than men, makecomments that are less evaluatively extreme and more neutral.

Analyses of the relationship between lies of liking and lies of dislikingrevealed a similar pattern of consistency in the cues used in perceptions ofdeception, along with a significant degree of inconsistency in the cues thatactudly do signal deceit. Thus, perceivers seem to assume that the samekinds of tell-tale cues are given off by liars feigning disliking as by liarsfeigning liking, when in fact the relationship between cues to liking lies andcue to disliking lies is substantially negative.

The cues examined in this study on the whole provide only a hint as towhat the most important verbal and nonverbal cues to actual and perceiveddeception might be. Half of the cues that we studied were not significantlyassociated with either perceived or actual deception. In a sense, the paucityof clear cues to actual deceit seems reasonable in light of the particular ex-perimental context. Subjects were lying as part of their role in fulfilling therequirements of the experiment; hence, guilt about lying should have beenminimal. Also, there were no explicit and compelling incentives for suc-cessful task performance, nor were there any disastrous consequences offailure; hence, anxiety should not have been a major variable either. Eventhe cognitive demands of fabricating a convincing lie should not have beenparticularly burdensome, since senders were allowed to take as much timeas they wished to plan each description before actually saying it out loud.Yet, in spite of this generally nonthreatening context, senders were not ablesuccessfully to conceal their deceit. Subjects in this study were able to detecteach other's deception (from audiovisual recordings) at an accuracysignificantly above chance (DePaulo & Rosenthal, 1979); other judges (whohad not served as senders in the study) were less accurate overall, but didsomewhat better at detecting deception when they did have access to verbalcues than when they did not (DePaulo, et al., 1982). Because of the spe-cial characteristics of this study (e.g., low motivation to lie, minimal guilt),results should be generalized with caution to other types of deceptive trans-actions. The available evidence (Zuckerman et al., 1981) suggests thatthe present study might wnrferestimate the degree to which verbal andnonverbal behaviors tip off deception, since more cues are reliablyassociated with deception when senders are highly motivated to lie suc-cessfully that when they are not highly motivated.

REFERENCE NOTE

1. Krauss, R. M., Geller, V., & Olson, C. Modalities and cues in the detection of deception.Paper presented at the meeting of the American Psychological Association, Washington,D.C, September 1976.


REFERENCES

Apple, W., Streeter, L. A., & Krauss, R. M. Effects of pitch and speech rate on personal attri-butions. Journal of Personality and Social Psychology, 1979, 37, 715-727.

Darwin, C. The expression of emotions in man and animals. London: John Murray, 1872.DePaulo, B. M., Lassiter, G. D., & Stone, J. I. Attentional determinants of success at de-

tecting deception and truth. Personality and Social Psychology Bulletin, 1982, 8, 213-219.DePaulo, B. M., & Rosenthal, R. Telling lies. Journal of Personality and Social Psychology,

1979, 37, 1713-1722.DePaulo, B. M., Rosenthal, R., Green, C. R., & Rosenkrantz, J. Diagnosing deceptive and

mixed messages from verbal and nonverbal cues. Journal of Experimental Social Psychol-ogy, 1982, 18, 433-446.

DePaulo, B. M., Zuckerman, M., & Rosenthal, R. Detecting deception: Modality effects.In L. Wheeler (Ed.), Review of personality and social psychology. New York: Sage Pub-lishers, 1980.

Ekman, P. Mistakes while deceiving. Annals of the New York Academy of Sciences, 1981,364, 269-278.

Ekman, P., & Friesen, W. V. Nonverbal leakage and clues to deception. Psychiatry, 1969,32, 88-106.

Ekman, P., Friesen, W. V., O'SulUvan, M., & Scherer, K. Relative importance of face, body,and speech in judgments of personality and affect. Journal of Personality and Social Psy-chology, 1980, 38, 270-277.

Freud, S. Collected papers. New York: Basic Books, 1959.Goldman-Eisler, F. Continuity of speech utterance, its determinants and its significance.

Language and Speech, 1961, 4, 220-224. (a)Goldman-Bisler, F. A comparative study of two hesitation phenomena. Language and

Speech, 1961, 4, 18-26. (b)Hocking, J. E., & Leathers, D. G. Nonverbal indicators of deception: A new theoretical

perspective. Communication Monographs, 1980, 47, 119-131.Kasl, S. V., & Mahl, G. F. The relationship of disturbances and hesitations in spontaneous

speech to anxiety. Journal of Personality and Social Psychology, 1965, 1, 425-433.Knapp, M. L., & Comadena, M. E. Telling it like it isn't: A review of theory and research on

deceptive communications. Human Communication Research, 1979, 5, 270-285.Krauss, R. M., Apple, W., Morency, N., Wenzel, C , & Winton, W. Verbal, vocal, and visi-

ble factors in judgments of another's affect. Joumal of Personality and Social Psychology,1981, 40, 312-320.

Kraut, R. E. Verbal and nonverbal cues in the perception of lying. Journal of Personality andSocial Psychology, 1978, 36, 380-391.

Kraut, R. B. Humans as lie detectors: Some second thoughts. Journal of Communication,1980, 30, 209-216.

Kraut, R. B., & Poe, D. Behavioral roots of person perception: The deception judgments ofcustoms inspectors and laymen. Journal of Personality and Social Psychology, 1980, 39,784-798.

Kuiken, D. Nonimmediate language style and inconsistency between private and expressedevaluations. Journal of Experimental Social Psychology, 1981, 17, 183-196.

Littlepage, G. E., & Pineault, M. A. Detection of deceptive factual statements from the bodyand the face. Personality and Social Psychology Bulletin, 1979, 5, 325-328.

Mahl, G. F. Disturbances and silences in the patient's speech in psychotherapy. Journal ofAbnormal and Social Psychology, 1956,53, 1-15.

Miller, N., Maruyama, G., Beaber, R. J., & Valone, K. Speed of speech and persuasion.Journal of Personality and Social Psychology, 1916,34, 615-624.


Peevers, B. H., & Secord, P. F. Developmental changes in attribution of descriptive con-cepts to persons. Journal of Personality and Social Psychology, 1973, 27, 120-128.

Slovic, P., & Lichtenstein, S. Comparison of the Bayesian and regression approaches to thestudy of information processing in judgment. Organizational Behavior and Human Per-formance, 1971, 6, 649-744.

Smith, B. L., Brovi'n, B. L., Strong, W. J., & Rencher, A. C. Effects of speech rate on per-sonality perception. Language and Speech, 1975, 18, 145-152.

Zuckerman, M., DePaulo, B. M., & Rosenthal, R. Verbal and nonverbal communication ofdeception. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 14).N.Y.: Academic Press, 1981.

Documents

Actual and Perceived Cues to Deception: A Closer Look at ...sarahita/SLP/depaulo.pdfPrevious research has shown both that speech can reliably reveal whether or not deception is occurring